Motion Compensation in Positron Emission Tomography ... · Respiration-averaged CT for attenuation correction in non-small-cell lung cancer. Eur J Nucl Med Mol Imaging. 2009 Apr;36(4):607-15

Institute of Radiopharmacy - PET centre ! Dr. Jens Langner ! www.fzd.de ! Member of the Leibniz Association

Motion Compensation inPositron Emission Tomography:

Performance of a Clinical Integration at the PET centre Dresden-Rossendorf

Jens LangnerPET centre - Institute of Radiopharmacy

Research Centre Dresden-Rossendorf, Germany

http://www.fzd.de

http://www.fzd.de

Institue of Radiopharmacy - PET centre ! Dr. Jens Langner ! www.fzd.de ! November 10, 2009 /38

Agenda

• introduction of the FZD and its PET facility

• motivation for working on motion compensation

• motion tracking and quantification of patient motion

• why list-mode? solved problems with ECAT scanners

• event-based motion compensation, its challanges and performance

• methods for clinical integration

• conclusion / summary

2


Introduction - the FZD

• Established: 1992

• Employees: ! 750,! 330 scientists (! 120 Ph.D. candidates)

• Budget: ! 61 Mio "

• Member of Leibniz Association

• 6 institutes and 6 large-scale facilities

Dresden

Ion Beam Centre

Radiation Source ELBE

Rossendorf Beamline

PET Centre

TOPFLOW Facility

Dresden High Magnetic Field Laboratory

Ion Beam Physics andMaterials Research

Radiation Physics

Radiochemistry

Radiopharmacy

Safety Research

High Magnetic Field Laboratory

Advanced Materials Research

Nuclear Safety Research

Cancer Research

3


Introduction - the FZD


• Employees: ! 750,! 330 scientists (! 120 Ph.D. candidates)

• Budget: ! 61 Mio "

• Member of Leibniz Association

• 6 institutes and 6 large-scale facilities

Dresden

Ion Beam Centre

Radiation Source ELBE

Rossendorf Beamline

PET Centre

TOPFLOW Facility

Dresden High Magnetic Field Laboratory

Ion Beam Physics andMaterials Research

Radiation Physics

Radiochemistry

Radiopharmacy

Safety Research

High Magnetic Field Laboratory

Advanced Materials Research

Nuclear Safety Research

Cancer Research

PETcentre

3


Introduction - PET centre

interdisciplinary group:

radiochemistrybiochemistrybiologymedicinecomputer sciencephysics

facilities:

cyclotronradiochemical laboratories (GMP)human + small animal PETsmall animal MRI and CThuman PET/MRI (in Q2/2010)

PET tracers:

16F-FDG, 18F-OMFD,18F-FDOPA, NaF, 11C-Acetate, 15O-water (GMP), ...


• In cooperation with the nuclear medicine department of the Technical University Dresden

• Head:Prof. Dr. Dr. Jörg van den Hoff

4



Research focus - PET group• radiotracer development

• preclinical and clinical PET studies in oncology and neurology

• characterization of selected parameters for PET-guided therapy response monitoring

• PET in radiation treatment planning

• quantitative evaluation of tracer kinetics

• multi-modality in vivo imaging

• PET in pharmaceutical research

5



Research focus - PET group• radiotracer development

• preclinical and clinical PET studies in oncology and neurology

• characterization of selected parameters for PET-guided therapy response monitoring

• PET in radiation treatment planning

• quantitative evaluation of tracer kinetics

• multi-modality in vivo imaging

• PET in pharmaceutical research

• improvement of PET imaging:

- optimized ROI delineation:e.g. automatic background dependent 3D-ROI delineation and segmentation

- utilization of list-mode data

- motion compensation techniques

5


Motivation - Motion Compensation

• PET is a functional method for imaging of biochemical and physiological processes in vivo

• provides absolute quantitative values for evaluation of metabolic processes

• current spatial resolution: ! 5 mm (brain), ! 8 mm (whole body)

• recent developments (e.g. PSF optimized reconstructions) demonstrate the feasibility of ! 2 mm

• acquisition times of several minutes to hours are unavoidable(low signal/noise ratio, dynamic acquisitions, etc.)

6



• PET is a functional method for imaging of biochemical and physiological processes in vivo

• provides absolute quantitative values for evaluation of metabolic processes

• current spatial resolution: ! 5 mm (brain), ! 8 mm (whole body)

• recent developments (e.g. PSF optimized reconstructions) demonstrate the feasibility of ! 2 mm

• acquisition times of several minutes to hours are unavoidable(low signal/noise ratio, dynamic acquisitions, etc.)

!patient movement increasingly limits the achievable spatial resolution in PET ...

6



... and immobilizations are of limited help only

7


Motivation - Consequences of patient motion

1. Intra-frame movement (dynamic + static acquisitions):

– loss in resolution („motion blurring“)

– reconstruction/image artifacts

– systematic errors in quantitative evaluations(e.g. standardized uptake value - SUV - in FDG whole body)

2. Inter-frame movement (dynamic acquisitions):

– incorrect registration between frames

– (massive) systematic errors in time-activity-curves (TAC)

– systematic errors in quantification of tracer kinetics

8


Motivation - Qualitative Consequences

Example: 18F-FDG PET

without motion with patient motion

0.0

[Bq/c

c]10280.0

9


Motivation - Quantitative Consequences

time-activity-curve (TAC)

Sagittal

Transaxial

Bq/c

c0.0

10290.0

0 5 10 15 20 25 30 35 40 45 50

10.000

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

Zeit [min]

Konzentr

ati

on [

Bq/cc]

expected progression (without motion)

male, 75 years18F-DOPA (M. Parkinson)248 MBq i.v.55 min dynamic emission(27 frames)

conce

ntr

atio

n [

Bq/c

c]

time [min]

10


Motivation - many approaches...1: Fin L, Bailly P, Daouk J, Meyer ME. Motion correction based on an appropriate system matrix for statistical reconstruction of respiratory-correlated PET acquisitions. Comput Methods Programs Biomed. 2009 May 30. [Epub ahead of print] PubMed PMID: 19487044.2: Costes N, Dagher A, Larcher K, Evans AC, Collins DL, Reilhac A. Motion correction of multi-frame PET data in neuroreceptor mapping: Simulation based validation. Neuroimage. 2009 May 27. [Epub ahead of print] PubMed PMID: 19481154.3: Wang X, Koch S. Positron emission tomography/computed tomography potential pitfalls and artifacts. Curr Probl Diagn Radiol. 2009 Jul-Aug;38(4):156-69. PubMed PMID: 19464586.4: Büther F, Dawood M, Stegger L, Wübbeling F, Schäfers M, Schober O, Schäfers KP. List mode-driven cardiac and respiratory gating in PET. J Nucl Med. 2009 May;50(5):674-81. Epub 2009 Apr 16. PubMed PMID: 19372491.5: Bai W, Brady M. Regularized B-spline deformable registration for respiratory motion correction in PET images. Phys Med Biol. 2009 May 7;54(9):2719-36. Epub 2009 Apr 8. PubMed PMID: 19351979.6: Nagamachi S, Wakamatsu H, Kiyohara S, Fujita S, Futami S, Arita H, Nishii R, Tamura S, Kawai K. Usefulness of a deep-inspiration breath-hold 18F-FDG PET/CT technique in diagnosing liver, bile duct, and pancreas tumors. Nucl Med Commun. 2009 May;30(5):326-32. PubMed PMID: 19282791.7: Yamaguchi T, Ueda O, Hara H, Sakai H, Kida T, Suzuki K, Adachi S, Ishii K. Usefulness of a breath-holding acquisition method in PET/CT for pulmonary lesions. Ann Nucl Med. 2009 Jan;23(1):65-71. Epub 2009 Feb 11. PubMed PMID: 19205840.8: Faber TL, Raghunath N, Tudorascu D, Votaw JR. Motion correction of PET brain images through deconvolution: I. Theoretical development and analysis in software simulations. Phys Med Biol. 2009 Feb 7;54(3):797-811. Epub 2009 Jan 9. PubMed PMID: 19131672.9: Raghunath N, Faber TL, Suryanarayanan S, Votaw JR. Motion correction of PET brain images through deconvolution: II. Practical implementation and algorithm optimization. Phys Med Biol. 2009 Feb 7;54(3):813-29. Epub 2009 Jan 9. PubMed PMID: 19131667.10: Hofmann M, Pichler B, Schälkopf B, Beyer T. Towards quantitative PET/MRI: a review of MR-based attenuation correction techniques. Eur J Nucl Med Mol Imaging. 2009 Mar;36 Suppl 1:S93-104. Review. PubMed PMID: 19104810.11: Daouk J, Fin L, Bailly P, Meyer ME. Respiratory-gated positron emission tomography and breath-hold computed tomography coupling to reduce the influence of respiratory motion: methodology and feasibility. Acta Radiol. 2009 Mar;50(2):144-55. PubMed PMID: 19096952.12: Cheng NM, Yu CT, Ho KC, Wu YC, Liu YC, Wang CW, Yen TC. Respiration-averaged CT for attenuation correction in non-small-cell lung cancer. Eur J Nucl Med Mol Imaging. 2009 Apr;36(4):607-15. Epub 2008 Dec 3. PubMed PMID: 19050875.13: Slomka PJ, Le Meunier L, Hayes SW, Acampa W, Oba M, Haemer GG, Berman DS, Germano G. Comparison of myocardial perfusion 82Rb PET performed with CT- and transmission CT-based attenuation correction. J Nucl Med. 2008 Dec;49(12):1992-8. PubMed PMID: 19038996.14: Lucignani G. Respiratory and cardiac motion correction with 4D PET imaging: shooting at moving targets. Eur J Nucl Med Mol Imaging. 2009 Feb;36(2):315-9. Review. PubMed PMID: 19030856.15: Bundschuh RA, Martinez-Müller A, Essler M, Nekolla SG, Ziegler SI, Schwaiger M. Local motion correction for lung tumours in PET/CT--first results. Eur J Nucl Med Mol Imaging. 2008 Nov;35(11):1981-8. Epub 2008 Aug 6. PubMed PMID: 18682940.16: Daouk J, Fin L, Bailly P, Meyer ME. Improved attenuation correction via appropriate selection of respiratory-correlated PET data. Comput Methods Programs Biomed. 2008 Oct;92(1):90-8. Epub 2008 Aug 3. PubMed PMID: 18676054.17: Fin L, Daouk J, Morvan J, Bailly P, El Esper I, Saidi L, Meyer ME. Initial clinical results for breath-hold CT-based processing of respiratory-gated PET acquisitions. Eur J Nucl Med Mol Imaging. 2008 Nov;35(11):1971-80. Epub 2008 Jun 26. PubMed PMID: 18581114.18: Janssen MH, Aerts HJ, Ollers MC, Bosmans G, Lee JA, Buijsen J, De Ruysscher D, Lambin P, Lammering G, Dekker AL. Tumor delineation based on time-activity curve differences assessed with dynamic fluorodeoxyglucose positron emission tomography-computed tomography in rectal cancer patients. Int J Radiat Oncol Biol Phys. 2009 Feb 1;73(2):456-65. Epub 2008 Jun 14. PubMed PMID: 18556143.19: Schäfers KP, Stegger L. Combined imaging of molecular function and morphology with PET/CT and SPECT/CT: image fusion and motion correction. Basic Res Cardiol. 2008 Mar;103(2):191-9. Review. PubMed PMID: 18324375.20: Alessio A, Kohlmyer S, Kinahan P. Consistency Driven Respiratory Phase Alignment and Motion Compensation in PET/CT. IEEE Nucl Sci Symp Conf Rec (1997). 2007;4:3115-3119. PubMed PMID: 19266056.

11


Motivation - ... but only a handful of paths

system matrixmodifications

withinreconstruction

multiple acquisition framing (MAF)

event-by-event correctionbeforereconstruction

listmode-based

CMS movementsinogram-based

co-registrationinter-frame

n/aintra-frameimage-based

Compensationtechniques

incorrect volumedelineation

TAC / SUV

quantitative effects

image artifacts

resolution loss (blurring)

Consequences

Patient motionin PET

12

! motion compensation for brain acquisitions in PET

• using the „raw data“ (list-mode) of a PET scanner

• using an external motion tracking device

• focused on clinical usability


Motivation - our approach

1) motion tracking and quantification of motion

2) routine acquisition of list-mode data

3) development and optimization of an event-based motion compensation algorithm

4) methods for integration into clinical routine(e.g. graphical user interfaces)

Tasks

13

• External motion tracking device(infrared video cameras - ARTtrack)

– spatial resolution better than 1 mm

– time resolution < 50 ms

– output of all six degrees of freedom (3x translations, 3x rotations)


Motion Tracking

Methods:

1.Installation/Calibration:e.g. calibration with PET coordinate system

2.Development and evaluation of a suitable motion target

3.Integration into clinical routine

14

http://www.ar-tracking.de/


Motion Tracking - Target development

• Problem: motion target has to output the true patient motion.

15

compare

teeth moulded target(as a reference)


Motion Tracking - Target development

• Problem: motion target has to output the true patient motion.

15

compare

teeth moulded target(as a reference)


Motion Tracking - Target evaluation results

16

I II III IV V

0.0 0.2 0.4 0.6 0.8 1.0

269.9

270.0

270.1

270.2

270.3

270.4

270.5

Euclidian difference

time [min]

tota

l [m

m]

min: 269.903 max: 270.489 diff: 0.586 mm

0.0 0.2 0.4 0.6 0.8 1.0

278

279

280

281

282

283

284

target 0 displacement vector length

time [min]

tota

l [m

m]

min: 279.515 max: 282.261 diff: 2.746 mm

0.0 0.2 0.4 0.6 0.8 1.0

327

328

329

330

331

332

333


time [min]

tota

l [m

m]

min: 327.120 max: 333.130 diff: 6.010 mm

Comparison between target 0 and target 1 (Tracking Coordinates)

Page 2/4test28.trk

Thu Aug 30 17:32:52 2007analyze_trk.R $Revision: 194 $ ! (c) 2005!2007 fzd.de

time [min]

dist

ance

[mm

]

�d = 0.6 mm

0.0 0.2 0.4 0.6 0.8 1.0

330.0

330.5

331.0

331.5

332.0


time [min]

tota

l [m

m]

min: 329.698 max: 331.960 diff: 2.262 mm

0.0 0.2 0.4 0.6 0.8 1.0

356

358

360

362

364

366

368


time [min]

tota

l [m

m]

min: 354.977 max: 369.313 diff: 14.336 mm

0.0 0.2 0.4 0.6 0.8 1.0

242

244

246

248

250

252

254

256


time [min]

tota

l [m

m]

min: 241.194 max: 255.374 diff: 14.180 mm


Page 2/4test21.trk


dist

ance

[mm

]

time [min]

�d = 2.3 mm

0.0 0.2 0.4 0.6 0.8 1.0

302

303

304

305

306


time [min]

tota

l [m

m]

min: 302.090 max: 306.303 diff: 4.213 mm

0.0 0.2 0.4 0.6 0.8 1.0

224

226

228

230

232

234

236


time [min]

tota

l [m

m]

min: 224.923 max: 234.322 diff: 9.398 mm

0.0 0.2 0.4 0.6 0.8 1.0

362

364

366

368

370

372

374

376


time [min]

tota

l [m

m]

min: 362.472 max: 375.965 diff: 13.493 mm


Page 2/4test18.trk


dist

ance

[mm

]

time [min]

�d = 4.2 mm

(a)

(b)

(c)

Euclidean distance plot


Motion Tracking - Target evaluation results

16

I II III IV V

0.0 0.2 0.4 0.6 0.8 1.0

269.9

270.0

270.1

270.2

270.3

270.4

270.5


time [min]

tota

l [m

m]

min: 269.903 max: 270.489 diff: 0.586 mm

0.0 0.2 0.4 0.6 0.8 1.0

278

279

280

281

282

283

284


time [min]

tota

l [m

m]

min: 279.515 max: 282.261 diff: 2.746 mm

0.0 0.2 0.4 0.6 0.8 1.0

327

328

329

330

331

332

333


time [min]

tota

l [m

m]

min: 327.120 max: 333.130 diff: 6.010 mm


Page 2/4test28.trk


time [min]

dist

ance

[mm

]

�d = 0.6 mm

0.0 0.2 0.4 0.6 0.8 1.0

330.0

330.5

331.0

331.5

332.0


time [min]

tota

l [m

m]

min: 329.698 max: 331.960 diff: 2.262 mm

0.0 0.2 0.4 0.6 0.8 1.0

356

358

360

362

364

366

368


time [min]

tota

l [m

m]

min: 354.977 max: 369.313 diff: 14.336 mm

0.0 0.2 0.4 0.6 0.8 1.0

242

244

246

248

250

252

254

256


time [min]

tota

l [m

m]

min: 241.194 max: 255.374 diff: 14.180 mm


Page 2/4test21.trk


dist

ance

[mm

]

time [min]

�d = 2.3 mm

0.0 0.2 0.4 0.6 0.8 1.0

302

303

304

305

306


time [min]

tota

l [m

m]

min: 302.090 max: 306.303 diff: 4.213 mm

0.0 0.2 0.4 0.6 0.8 1.0

224

226

228

230

232

234

236


time [min]

tota

l [m

m]

min: 224.923 max: 234.322 diff: 9.398 mm

0.0 0.2 0.4 0.6 0.8 1.0

362

364

366

368

370

372

374

376


time [min]

tota

l [m

m]

min: 362.472 max: 375.965 diff: 13.493 mm


Page 2/4test18.trk


dist

ance

[mm

]

time [min]

�d = 4.2 mm

(a)

(b)

(c)

Euclidean distance plot


Motion Quantification - ambiguity problem

• Problem: drawing conclusions on the significance of motion from the six degrees of freedom can be ambiguous

17

http://dict.leo.org/ende?lp=ende&p=5tY9AA&search=ambiguity





17

" " "






17

? ? ?

" " "




Motion Quantification - Method

Method:

1.apply the motion parameters to selected points on an imaginary spherewithin the FOV (sphere approximates a human head ! 20 cm diameter)

2.calculate the euclidean distance for each point

18

(a)

x

y

z

(b)


Motion Quantification - Example

19

!10 0 10 20 30 40 50

05

10

15

20

acquisition time [min]

mo

ve

me

nt

[mm

]

10 sign.motion found on artifical head surface (r=100mm)3 sign.motion found in mean striatum areas

"best" point on head surface

"worst" point on head surface

L striatum [p1=56:60:36]

R striatum [p2=78:60:36]

point of motion > 4 mm

frame boundaries

accessible range on head surface

"best" point on head surface (10%!90%) = 0.8!7.5 mm

minimum movement [mm]

Fre

qu

en

cy

0 2 4 6 8

02

00

04

00

0

"worst" point on head surface (10%!90%) = 0.8!17.7 mm

maximum movement [mm]

Fre

qu

en

cy

0 5 10 15 20

02

00

04

00

06

00

0

sphere angle !! [deg]

sp

he

re a

ng

le ""

[d

eg

]

0 45 90 135 180 225 270 315 360

04

59

01

35

18

0

Neck

Os temporale (L)

Os parietale

Neck

Os nasale

Os parietale

Neck

Os temporale (R)

Os parietale

Neck

Os occipitale

Os parietale

distance fluctuation map (sd.) ! max. = +/!7.3 mm

01

23

45

67


sp

he

re a

ng

le ""

[d

eg

]

0 45 90 135 180 225 270 315 360

04

59

01

35

18

0

Neck

Os temporale (L)

Os parietale

Neck

Os nasale

Os parietale

Neck

Os temporale (R)

Os parietale

Neck

Os occipitale

Os parietale

mean distance to source point map ! max. = 10.1 mm

02

46

81

0

Patient head movement analysis with significant motion (> 4 mm) selected:

Page 2/3./hackbarth_070911.tfm

Tue Sep 11 13:58:11 2007analyze_trk.R $Revision: 195 $ ! (c) 2005!2007 fzd.de


Motion Quantification - Example

20

40

50

60

Neck

70

80

0350

340

10

3020

40 50 60 80709090

100

150

140

130

110120

160

170

180

190

200

300

310

320

330

100

Os temporale (L)

110

120

Os nasale

130

Min point

Max point

140

150

160

170180

Os parietale

Os temporale (R)Os occipitale

290

280

270

260250

240 220230

210

0

10

20

30

Neck

50

60

70

80

10

20

30

0

350

40

5060

100

110

120

130

150

140

160

170

190

180

200210

220230 240 250 260 270

280290

300

310

320

340

330Os temporale (R)

Os occipitale

Os parietale

180

170

160

150

Os temporale (L)

140

Max point

Min point

130

120

Os nasale

110

100

9090

8070

304020

10

0

(a)

3D distanceFluctuation sphere plot

!10

010

20

30

40

50

0 5 10 15 20

acquis

ition tim

e [m

in]

movement [mm]

10

sig

n.m

otio

n fo

un

d o

n a

rtifica

l he

ad

su

rfac

e (r=

10

0m

m)

3 s

ign

.mo

tion

fou

nd

in m

ea

n s

triatu

m a

rea

s

"be

st" p

oin

t on

he

ad

su

rface

"wo

rst" p

oin

t on

he

ad

su

rface

L s

triatu

m [p

1=

56

:60

:36

]

R s

triatu

m [p

2=

78

:60

:36

]

po

int o

f mo

tion

> 4

mm

fram

e b

ou

nd

arie

s

acce

ssib

le ra

ng

e o

n h

ea

d s

urfa

ce

"best" p

oin

t on

head

su

rface (1

0%!

90%

) = 0

.8!

7.5

mm

min

imum

movem

ent [m

m]

Frequency

02

46

8

0 2000 4000

"wo

rst" p

oin

t on

head

su

rface (1

0%!

90%

) = 0

.8!

17.7

mm

maxim

um

movem

ent [m

m]

Frequency

05

10

15

20

0 2000 4000 6000

sphere

angle

!! [d

eg]

sphere angle "" [deg]

045

90

135

180

225

270

315

360

0 45 90 135 180

Ne

ck

Os te

mp

ora

le (L

)

Os p

arie

tale

Ne

ck

Os n

asa

le

Os p

arie

tale

Ne

ck

Os te

mp

ora

le (R

)

Os p

arie

tale

Ne

ck

Os o

ccip

itale

Os p

arie

tale

dis

tan

ce

fluc

tua

tion

ma

p (s

d.) !

ma

x. =

+/!

7.3

mm

0 1 2 3 4 5 6 7

sphere

angle

!! [d

eg]


045

90

135

180

225

270

315

360

0 45 90 135 180

Ne

ck

Os te

mp

ora

le (L

)

Os p

arie

tale

Ne

ck

Os n

asa

le

Os p

arie

tale

Ne

ck

Os te

mp

ora

le (R

)

Os p

arie

tale

Ne

ck

Os o

ccip

itale

Os p

arie

tale

me

an

dis

tan

ce

to s

ou

rce

po

int m

ap

! m

ax

. = 1

0.1

mm

0 2 4 6 8 10

Patie

nt h

ead m

ovem

ent a

naly

sis

with

sig

nific

ant m

otio

n (>

4 m

m) s

ele

cte

d:

Page 2

/3./h

ackbarth

_070911.tfm

Tue S

ep 1

1 1

3:5

8:1

1 2

007

an

aly

ze

_trk

.R $

Re

vis

ion

: 19

5 $

! (c

) 20

05!

20

07

fzd

.de

[mm]

(b)


List-mode: Measuring principle of PET

21

CoincidenceProcessing Unit

Image ReconstructionAnnihilation

Coincidence Data(Sinogram/List-mode)

e+

e−

γ

γ

ν

• Registration of coincidences (events) between two detectorsi.e. Line-of-Response (LOR)


PET acquisition - why list-mode?

• highest time and data resolution possible(raw data vs. compression of LORs with sinograms)

• allows to retrospectively change the framing scheme

• modification of coincidence data prior to image reconstruction(e.g. apply motion correction, list-mode based reconstructions, etc.)

22

“framing and histograming“(sinogram generation)

time coincidence events (“list mode stream”)

… …

1 2 3 4 . . . . . n-1. n n+1 . . N

image reconstruction

… …

time frames

data storage(hard disk)


List-mode problems with ECAT scanners (HR+)

• Direct access to list-mode data often limited or not supported at all(! 0.5 MB/s with an ECAT EXACT HR+)

! no clinical usability due to long transfer timese.g. for a typical amount of 4-6 GB ! 3 hours

0.5 MB/s

acquisition system (ACS)


10 MB/s

[1] Langner, J. et. al; Z. Med. Phys. 2006; 16(1):75-82.

23

PETgantry

Method A:

70 MB/s

SCSI-RAID(dual channel)

LinuxSystem



• Direct access to list-mode data often limited or not supported at all(! 0.5 MB/s with an ECAT EXACT HR+)

! no clinical usability due to long transfer timese.g. for a typical amount of 4-6 GB ! 3 hours

0.5 MB/s

acquisition system (ACS)


10 MB/s

1 GBit/s

• optimised data access by providing a different data access path(! 70 MB/s using dual-channel SCSI-RAID system)

! allows transfer < 1 min [1] Langner, J. et. al; Z. Med. Phys. 2006; 16(1):75-82.

23

PETgantry

Method A:

HD

fiber-optic

ACS2

M1M2

M3 opto-electronicadapter

Linux

HDDAQ

externaldata bus

DAQcards

ACS2data bus

optocoupler

powersource



24

Method B:


adapter card

HD

fiber-optic

ACS2

M1M2

M3 opto-electronicadapter

Linux

HDDAQ

externaldata bus

DAQcards

ACS2data bus

optocoupler

powersource



24

Method B:


• optocoupler-based adapter card to read out the list-mode data directly from an external data bus of ACS

! receive coincidences in real-time during acquisition

! apply motion correction or image reconstruction in real-time

adapter card


Event-based motion correction

• Method:

1.spatial transformation of all registered events

2.sorting events into sinograms

3.apply standard reconstruction(e.g. OSEM)

• complications:

1. detector normalisation

2. LOR discretisation

3. Out-of-FOV correctionschematic view

25

[1] Bühler, P., et. al; IEEE T MED IMAGING 23(9) 2004



• Method:




• complications:




25




• Method:




• complications:




25




D1

D2

• Method:




• complications:




25




D1

D2

• Method:




• complications:




25




D1

D2

D1´

D2´

• Method:




• complications:




25


• „Loss“ of events due to the transformation of LORs outside the field-of-view (FOV)


Out-of-FOV correction

26

schematic view




26

schematic view 30s




26

schematic view 30s




26

schematic view 30s30s




26

schematic view 30s30s

f =Acqtime

Acqtime−Out-of-FOV-time

Ncorrected = Nmeasured · f


• Initial approach [1]:




26

schematic view 30s

! This approach does not always solve the problem:

Sagittal

0,0

5986

Bq/c

c

30s

f =Acqtime

Acqtime−Out-of-FOV-time

Ncorrected = Nmeasured · f


• Initial approach [1]:


Optimisation Out-of-FOV correction

27


posi

tion [

mm

]

discrete patient motion (example)

Method:

Goal: minimisation of Out-of-FOV factors

• Original approach:

transformation of all events to acquisition start (tref =0)

• Optimised approach:

1. Analysis of motion data

2. Identify optimal reference time(tref = topt)

• condition: position in which the patient has been most of the time

3. Transformation of all events totime tref



27


posi

tion [

mm

]


tref = 0

Method:











27


posi

tion [

mm

]


topttref = 0

Method:











27

tref =


posi

tion [

mm

]


topt

Method:










Results of optimized Out-of-FOV correction

28

1 1 2 2 3 3

15

1050

100

500

1000

data set

fact

or

4 4 5 5 6 6

!

!

!

!

!

!

!

!

!

!

!

!

!!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!!

!

!

!

!

!

!

!!

!

!

!

!

!!!!

!

!

!

!

!

!

!

!

!

!

!

!!

!!

!!

!

!

!

!

!

!

!

!!

!

!

!!!

!

!!

!

!

!

!!

!!

!

!

!

!

!!

!

!

!

!

!

!!

!

!!!!

!!

!

!

!

!!

!

!

!!!

!

!

!

!!!

!!!

!!!

!

!

!!

!

!

!

!

!

!

!

!

!!

!

!

!!

!!

!

!

!!!

!

!!

!!!

!

!

!

!

!!

!

!!

!

!

!

!

!!

!

!!!

!

!!!

!

!!!

!

!!

!

!!

!

!

!

!

!!

!

!

!

!

!

!!!

!

!!

!

!

!!

!

!

!

!

!

!

!!

!!

!

!

!!

!

!!!!!

!

!

!

!!

!

!

!!

!

!

!

!!

!

!

!

!

!

!

!

!!

!

!

!

!

!

!

!

!

!

!

!!

!

!

!!

!

!!

!

!

!

!

!

!

!

!

!

!

!

!!

!

!

!

!

!!

!

!!!

!!

!

!!!

!

!!!!!

!

!!!!!!!

!

!

!

!

!

!

!

!

!!

!

!

!

!

!

!

!

!!

!

!

!

!

!!

!

!

!!

!

!!

!

!

!!!!!

!!!

!

!!

!

!

!

!

!

!

!!

!!

!!!

!

!!!

!

!!

!

!

!

!

!!

!

!

!!

!

!

!

!

!

!

!

!

!

!

!!

!!!

!!

!

!

!

!!

!!

!

!

!!!!!

!

!

!!

!!

!

!

!!!

!!

!

!

!

!!!!

!!!

!

!

!

!

!!

!

!!

!!!!!

!

!!

!

!

!

!

!!

!

!!

!

!

!

!

!

!

!!!!!!

!

!!!

!

!

!

!

1

15

10

50

100

500

1000

OFOV factors

study

facto

r

!!

!!!!

!!!!!!!!!

!!!

!!

!!!!

!!

!!

!!!!!!

!!!!!!!!!!

!!!

!!

!!!!!!

!!!!!!!!!!!!

!!!!!

!!

!!!

!

!

!

!!!

!!!!!!!!

!!!

!

!!

!

!!!

!

!

!!!!!!!!!!!!!!!

!

!!

!!

!

!!!

!!

!!!!!!!

!!!!

!

!!!!!!!

!

!

!!

!

!

!

!!

!!

!

!!!!!

!!

!

!!!!

!

!

!!!

!

!

!!!

!!!!

!

!!

!!

!

!!

!

!

!

!

!

!

!!

!!!

!

!!

!

!

!!!!

!!

!!!

!!

!

!!!

!!!!!!!

!

!!

!

!!!!

!

!!

!

!!

!

!!

!

!

!!!

!

!!

!!!!!

!

!

!

!!

!

!

!!!!!

!

!!!

!

!!

!

!!

!!

!

!!!

!!!

!

!!!!

!!!!!!!

!!!

!

!

!

!!

!!

!

!

!!!!!!

!!!!!!

!

!!!!!

!!!!

!!

!

!

!

!!!!!!!

!

!!

!

!

!!!!!!!

!!

!!

!!!

!

!

!

!

!

!!!!!!!!!!!!

!!!!!!!

!

!!

!

!!

!

!

!

!!!!!

!!

!!

!!

!!!!

!

!!

!!!!

!!!!!!

!

!

!!

!!

!!!!!

!!

!!!

!!!

!

!!

!

!!!

!!!!

!!

!!!!!!

!!!!!!!

!!!!

!

!!!!!!!

!!

!!!!!!!!!!!!!

!!!

!!!!

!

!

!

!!

!!!!!!!!!!!!!!!!

!!

!!!!!!!!!!!!!!!!!!!!!!!!!!

!!!!!!!!!!!!!!!!!!!!!!!!!!!

!!!!!!!!!

!

!!!

!

!

!!!

!!!!!!!

!

!

!!!

!!

!!

!

!

!

!!!!!!!!!!!!!

!

!!!!!!!

!!

!

!

!!

!!!

!!

!

!!!!

!!!!!!!

!!!!!!!!!!!

!!!!

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

!!!!!

!!!!

!!!!!

!

!!!!!!

!!!!

!

!

!

!

!!!

!

!!

!

!!

!!

!

!!

!

!!!!!!!!!!!!!!!!!!!!!!!!!

!!!!!!!!!!!!!!!

!!!

!!!!!!!

!!!!!!!!!!!!!!!!!!!!

!!

!!!!!!!!

!!!!!!!!!!!!!!!!

!!

!!!!!!!!!!

!!!!!

!!!!!!!!!!!!!

!!!!!!!!

!

!

!

!

!

!!

!!!!!!

!!!!!!!!!!!!!!!!

!!

!!

!

!

!!

!!!!!!

!!!!!!!!

!

!!!!!!

!!!!!!

!

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

!

!!!!!!!!!

!!!!!!

!

!

!

!!!!!

!

!!!!

!

!!

!

!

!!!

!!

!!!!!!

!!!!

!

!

!!

!!

!!

!!

!

!!!!!!!!!

!

!!

!

!!!!!

!!

!!!!!!!!

!!!!!!!!!!!!

!!!!!!

!!!

!

!

!

!

!

!!!!!

!!

!!

!

!!!!

!!!!

!!!!!!!

!

!

!

!

!

!!!!

!!

!!

!

!

!!!!

!!

!!!!!

!

!!!!

!

!

!!!!!!!!!!!!!!!!!!!

!!

!

!

!

!!!!!!!!!!

!!!

!

!!!!

!!!!

!

!

!!!

!!!!!

!!!!

!

!!!!

!

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

!!

!!

!!!!!!

!

!

!!!!!!

!!!!

!!!

!!!!!!!!

!!!!!!!!!!!!!!!!!

!!!!!!

!!!!!!!!!!!!!!!!!

!

!!

!!

!!!!

!

!!!!

!!!!!!!!

!!!!!!!!!!

!!!

!!!

!!

!

!!!

!

!

!!!!!!

!

!!!

!

!!

!

!!!!!!!!!

!

!!!!!!!!!!!!!!!

!

!!

!!!!!!!!!!!!

!

!!

!

!!!!!!!!

!!

!

!!

!

!!

!!!!!!!!!!!!!!!!!!!!!

!!!!!!!!

!

!!!!!!

!

!!!!!!!!

!!!

!!!!!!!!!!!!!

!!!!!!!!!!!

1

15

10

50

100

500

1000

!

!

!!!

!

!

!

!!!!

!!

!

!

!

!

!

!

!

!

!

!

!

!!

!

!!

!!!

!!!

!

!!!!

!!!

!

!!!

!

!

!

!

2

15

10

50

100

500

1000

!!!

!!!!!!!

!!!!!!

!!!!!!!!!!

!!!

!!!!!

!!!!!!!!!!!!!

!!!

2

15

10

50

100

500

1000

!

!

!

!!!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!!

!

!

3

15

10

50

100

500

1000

3

15

10

50

100

500

1000

!

!

!

!

!!

!

!!

4

15

10

50

100

500

1000

!!!!!!!!!!!!!!!!

4

15

10

50

100

500

1000

!!

5

15

10

50

100

500

1000

!

5

15

10

50

100

500

1000

!!

!

!

6

15

10

50

100

500

1000

!!

!

!

!!

!

6

15

10

50

100

500

1000

non!optimised OFOV correction

optimised OFOV correction

Distribution of Out-of-FOV factorsO

FOV fac

tor

Distribution of Out-of-FOV factors (boxplot)

data set

non-optimisedoptimised

• Noticable reduction of Out-of-FOV factors

• Noticable reduction of image artefacts


Results of optimized Out-of-FOV correction

28

(a)

(b)

059

81Bq

/ml

Transaxial Coronal Sagittal

Qualitative Comparisonqualitative comparison

withoutOptimisation

withOptimisation

• Noticable reduction of Out-of-FOV factors

• Noticable reduction of image artefacts


Results of MC - point source evaluations

29

At rest(reference image)

WithoutMotion Correction

WithMotion Correction

Qualitative Comparison

[MBq

/ml]

0.00

2.37t: 5-10 min t: 5-10 mint: 0-5 min

5 mm


Results of MC - example patient data I

withoutmotion correction

male, 64 yearsdifferential diagnosesM. Parkinson

PET acquisition:F-18 DOPA171 MBq i.v.10 min transmission55 min emission(27 frames)

detected motion:1,2 - 19,4 mmmean: 11,3 mm

conce

ntr

atio

n [

Bq/c

c]0.0

4344.6

7

• better delineation of the striatum

• reduction of image artefacts

Transaxial Transaxial

30

withmotion correction



31

Quantitative evaluation via a tracer kinetics analysis:

• standard procedure at our PET centre for M. Parkinson evaluation:

1.positioning of 8 region-of-interest (ROI) in the striatum+ 1 ROI in the occipital lobe as a reference tissue

2.comparison of time-activity-curves (TAC); calculation of the tracer kinetics parameter R0k3 on the base of an irreversible reference tissue two compartment model (Patlak plot)

ncrpra

prmprpref

nclpla

plpplm

|�a|>| �b|>|�c|>| �d|>|�e|�

�� =| �a �|>| �e �|>| �d �|>| �c �|>| �b �|

��

�|�a|>| �b|>|�c|>| �d|>|�e|�

�� =| �a �|>| �e �|>| �d �|>| �c �|>| �b �|

��

�

ncr nucleus caudatus right

nucleus caudatus leftnclpra putamen right anterior

prm putamen right mediusprp putamen right posterior

pla putamen left anterior

plm putamen left medius

plp putamen left posterior

ref reference ROI

pr

pl

quantitative evaluation



32

• relevant change of R0k3

! e.g. ! 230 % in ncl

without MC

ROI volume [cc]

R0k 3

mea

n [

1/m

in]

with MC

0.52 0.52 1.6 1.6 0.52 0.52 0.52 0.52 0.52 0.52

original_roi_patlak.txt

Volume (0.524445)

R0

k3

0.0

00

0.0

05

0.0

10

0.0

15

roi.n

cr

roi.n

cl

roi.p

r

roi.p

l

roi.p

ra

roi.p

rm

roi.p

rp

roi.p

la

roi.p

lm

roi.p

lp

0.0156

0.0046

0.0019

0.0041 0.0042

0.0032

0.0022

0.0095

!0.0008

0.0013

0.52 0.52 1.6 1.6 0.52 0.52 0.52 0.52 0.52 0.52

corrected_roi_patlak.txt

Volume (0.524445)

R0

k3

0.0

00

0.0

05

0.0

10

0.0

15

roi.n

cr

roi.n

cl

roi.p

r

roi.p

l

roi.p

ra

roi.p

rm

roi.p

rp

roi.p

la

roi.p

lm

roi.p

lp

0.0127

0.0105

0.0095 0.0095 0.0094

0.0110

0.0088 0.0089

0.0080

0.0102

ncr

ncl

pr

pl

0.52 0.52 1.6 1.6 0.52 0.52 0.52 0.52 0.52 0.52

original_roi_patlak.txt

Volume (0.524445)

R0

k3

0.0

00

0.0

05

0.0

10

0.0

15

roi.n

cr

roi.n

cl

roi.p

r

roi.p

l

roi.p

ra

roi.p

rm

roi.p

rp

roi.p

la

roi.p

lm

roi.p

lp

0.0156

0.0046

0.0019

0.0041 0.0042

0.0032

0.0022

0.0095

!0.0008

0.0013

0.52 0.52 1.6 1.6 0.52 0.52 0.52 0.52 0.52 0.52

corrected_roi_patlak.txt

Volume (0.524445)

R0

k3

0.0

00

0.0

05

0.0

10

0.0

15

roi.n

cr

roi.n

cl

roi.p

r

roi.p

l

roi.p

ra

roi.p

rm

roi.p

rp

roi.p

la

roi.p

lm

roi.p

lp

0.0127

0.0105

0.0095 0.0095 0.0094

0.0110

0.0088 0.0089

0.0080

0.0102

ncr

ncl

pr

pl

ncr

pr

ref

ncl

pl|�a|>| �b|>|�c|>| �d|>|�e|�

�� =| �a �|>| �e �|>| �d �|>| �c �|>| �b �|

��

�|�a|>|� b|>|�c|>|� d|>|�e|

��=|� a� |>|� e� |>|� d� |>|� c� |>|� b� |

��


Results of MC - example patient data II

time-activity-curve

Bq/c

c0.0

10290.0

0 5 10 15 20 25 30 35 40 45 50

10.000

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

Zeit [min]

Konzentr

ati

on [

Bq/cc]

with MC

without MC Transaxial

male, 75 yearsF-18 DOPA (M. Parkinson)248 MBq i.v.55 min emission(27 frames)

33

time [min]

conce

ntr

atio

n [

Bq/c

c]


Clinical Integration - Graphical User Interfaces

(a) (b)

34

!10 0 10 20 30 40 50

05

10

15

20


movem

ent [m

m]







frame boundaries




Fre

quency

0 2 4 6 8

02000

4000



Fre

quency

0 5 10 15 20

02000

4000

6000


sphere

angle

"" [deg]

0 45 90 135 180 225 270 315 360

045

90

135

180

Neck

Os temporale (L)

Os parietale

Neck

Os nasale

Os parietale

Neck

Os temporale (R)

Os parietale

Neck

Os occipitale

Os parietale


01

23

45

67


sphere

angle

"" [deg]

0 45 90 135 180 225 270 315 360

045

90

135

180

Neck

Os temporale (L)

Os parietale

Neck

Os nasale

Os parietale

Neck

Os temporale (R)

Os parietale

Neck

Os occipitale

Os parietale


02

46

810





Clinical Integration - Motion report pages

35

Analysed subject:Acquisition time:Total # motions:

Cross!Calibration:

./hackbarth_070911.trk55.00 min42674 (a 77 ms)cc_070901.tfm

X Y ZReference target motion:(should be ~0.0)

0.059 0.023 0.026 [mm]0.001 0.004 0.004 [deg]

X Y ZPatient target motion: 2.554 11.007 12.657 [mm]

5.162 1.373 1.783 [deg]

Head movement analysis on head surface (r=100mm):

'best' point on head surface:

'worst' point on head surface:

distance fluctuation:

mean distance:

10 sign. motions > 4 mm0.8!7.5 mm

0.8!17.7 mm

max. +/! 7.3 mm

max. 10.1 mm

on 'mean striatum' areas:L striatum [56:60:36]:

R striatum [78:60:36]:


0.0!14.4 mm

Estimated Score:(1: no motion, 2: low motion, 3: moderate motion,

4: significant motion, 5: high motion)5

Page 1/3

./hackbarth_070911.trk

Tue Sep 11 13:58:11 2007

analyze_trk.R $Revision: 195 $ ! (c) 2005!2007 fzd.de

Motion!Tracking SummaryTue Sep 11 13:58:11 2007


Cross!Calibration:



0.059 0.023 0.026 [mm]0.001 0.004 0.004 [deg]


5.162 1.373 1.783 [deg]





mean distance:


0.8!17.7 mm

max. +/! 7.3 mm

max. 10.1 mm




0.0!14.4 mm



Page 1/3


Tue Sep 11 13:58:11 2007




Cross!Calibration:



0.059 0.023 0.026 [mm]0.001 0.004 0.004 [deg]


5.162 1.373 1.783 [deg]





mean distance:


0.8!17.7 mm

max. +/! 7.3 mm

max. 10.1 mm




0.0!14.4 mm



Page 1/3


Tue Sep 11 13:58:11 2007



(a)

(b)

(c)

(d)

doe_070911.trk

40

50

60

Neck

70

80

0350

340

10

3020

40 50 60 80709090

100

150

140

130

110120

160

170

180

190

200

300

310

320

330

100

Os temporale (L)

110

120

Os nasale

130

Min point

Max point

140

150

160

170180

Os parietale

Os temporale (R)Os occipitale

290

280

270

260250

240 220230

210

0

10

20

30

Neck

50

60

70

80

10

20

30

0

350

40

5060

100

110

120

130

150

140

160

170

190

180

200210

220230 240 250 260 270

280290

300

310

320

340

330Os temporale (R)

Os occipitale

Os parietale

180

170

160

150

Os temporale (L)

140

Max point

Min point

130

120

Os nasale

110

100

9090

8070

304020

10

0

!10 0 10 20 30 40 50

05

10

15

20


mo

ve

me

nt

[mm

]







frame boundaries




Fre

qu

en

cy

0 2 4 6 8

02

00

04

00

0



Fre

qu

en

cy

0 5 10 15 20

02

00

04

00

06

00

0


sp

he

re a

ng

le ""

[d

eg

]

0 45 90 135 180 225 270 315 3600

45

90

13

51

80

Neck

Os temporale (L)

Os parietale

Neck

Os nasale

Os parietale

Neck

Os temporale (R)

Os parietale

Neck

Os occipitale

Os parietale


01

23

45

67


sp

he

re a

ng

le ""

[d

eg

]

0 45 90 135 180 225 270 315 360

04

59

01

35

18

0

Neck

Os temporale (L)

Os parietale

Neck

Os nasale

Os parietale

Neck

Os temporale (R)

Os parietale

Neck

Os occipitale

Os parietale


02

46

81

0




(a)

3D distanceFluctuation sphere plot

!10

010

20

30

40

50

0 5 10 15 20

acquis

ition tim

e [m

in]

movement [mm]

10 s

ign

.mo

tion

fou

nd

on

artific

al h

ead

su

rface (r=

100m

m)

3 s

ign

.mo

tion

fou

nd

in m

ean

stria

tum

are

as

"best" p

oin

t on h

ead s

urfa

ce

"wors

t" poin

t on h

ead s

urfa

ce

L s

triatu

m [p

1=

56:6

0:3

6]

R s

triatu

m [p

2=

78:6

0:3

6]

poin

t of m

otio

n >

4 m

m

fram

e b

oundarie

s

accessib

le ra

nge o

n h

ead s

urfa

ce

"best" p

oin

t on

head

su

rface (1

0%!

90%

) = 0

.8!

7.5

mm

min

imum

movem

ent [m

m]

Frequency

02

46

80 2000 4000

"wo

rst" p

oin

t on

head

su

rface (1

0%!

90%

) = 0

.8!

17.7

mm

maxim

um

movem

ent [m

m]

Frequency

05

10

15

20

0 2000 4000 6000

sphere

angle

!! [d

eg]


045

90

135

180

225

270

315

360

0 45 90 135 180

Neck

Os te

mpora

le (L

)

Os p

arie

tale

Neck

Os n

asale

Os p

arie

tale

Neck

Os te

mpora

le (R

)

Os p

arie

tale

Neck

Os o

ccip

itale

Os p

arie

tale

dis

tan

ce flu

ctu

atio

n m

ap

(sd

.) ! m

ax. =

+/!

7.3

mm

0 1 2 3 4 5 6 7

sphere

angle

!! [d

eg]


045

90

135

180

225

270

315

360

0 45 90 135 180

Neck

Os te

mpora

le (L

)

Os p

arie

tale

Neck

Os n

asale

Os p

arie

tale

Neck

Os te

mpora

le (R

)

Os p

arie

tale

Neck

Os o

ccip

itale

Os p

arie

tale

mean

dis

tan

ce to

so

urc

e p

oin

t map

! m

ax. =

10.1

mm

0 2 4 6 8 10

Pa

tien

t he

ad

mo

ve

me

nt a

na

lysis

with

sig

nific

an

t mo

tion

(> 4

mm

) se

lecte

d:

Page 2

/3./h

ackbarth

_070911.tfm

Tue S

ep 1

1 1

3:5

8:1

1 2

007

an

aly

ze

_trk

.R $

Re

vis

ion

: 19

5 $

! (c

) 20

05!

20

07

fzd

.de

[mm]

(b)

(c)

(d)

• patient motion has a potentially high influence on the image quality and on the accuracy of a tracer kinetics analysis

• a quantitative motion compensation is possible, can be integrated into clinical routine and helps minimising or even eliminating motion related effects

• the quantification of patient motion via an external motion tracking device is independent from applying a motion correction and provides additional information for the evaluation of questionable image artifacts

• a motion correction can be an important factor for the possibility to perform certain studies (e.g. in case of long scan times)

• the amount of „lost scans“ due to patient motion can be reduced in clinical routine

• parts of the methods have already been transfered to another centre(Forschungszentrum Jülich, Germany)


Conclusion / Summary

36


Patient study underway...

0-3 mm34%

3-5 mm31%

5-7 mm18%

> 7 mm18%

• 912 clinically acquired brain scans incl. motion correction performed until today

– 666 F18-DOPA (M. Parkinson)

– 143 F18-FDG (M. Alzheimer)

– 54 F18-OMFD (oncology)

– 49 others

• ToDo:

1)Evaluation of the image quality improvement

2)Selection of suitable patient data for detailed analysis

3)Evaluation of the changes in tracer kinetics due to the motion compensationmaximum motion

detected

37

• Jörg van den Hoff

• Paul Bühler

• Frank Hofheinz

• Jens Langner

• Uwe Just

• Hagen Mölle

• Christian Pötzsch

• Edmund Will

• Sören Dittrich


Who made it possible?

Danke für Ihre

Aufmerksamkeit38

• Bettina Beuthien-Baumann

• Liane Oehme

... and the city of Dresden

Documents

Motion Compensation in Positron Emission Tomography ... · Respiration-averaged CT for attenuation correction in non-small-cell lung cancer. Eur J Nucl Med Mol Imaging. 2009 Apr;36(4):607-15